This server has a single tool 'run' with critical definition and design flaws. The tool name violates naming conventions (generic 'run' instead of verb_noun like 'search_failures' or 'analyze_failures'), the description is vague about actual capabilities, parameter descriptions lack essential context, and the schema is minimally structured. Input validation is present but error messages are not actionable. Output structure is underdocumented. The server treats a complex multi-faceted operation (search + ranking + failure extraction + replay evaluation) as a single monolithic tool, violating the single-responsibility principle.

• Rename tool from 'run' to 'search_workflow_failures' or 'analyze_failure_records'. Start with a clear action verb.
• Expand tool description to ~100 chars following pattern: 'Searches workflow failure records by query terms, ranks matches by relevance score, and filters by status. Returns ranked results with evidence excerpts and a list of records marked as failed or errored. Use this to find relevant failures for replay diagnosis.'
• Document input schema explicitly with JSON Schema: { type: 'object', properties: { records: { type: 'array', items: { type: 'object', description: 'Workflow record with required fields: status (string), title (string), body (string), and optional fields (id, timestamp, etc).' }, description: 'List of workflow records to search' }, query: { type: 'string', minLength: 1, maxLength: 500, description: 'Search query: one or more lowercase terms separated by spaces (e.g. "timeout error"). Terms are matched case-insensitively against record content. Special characters and boolean operators are not supported.' } } }
• Document output schema: { type: 'object', properties: { status: { type: 'string', enum: ['ok', 'error'], description: 'Success indicator' }, query: { type: 'string', description: 'Original query' }, matches: { type: 'array', items: { type: 'object', properties: { rank: { type: 'integer', description: 'Result rank (1 = highest relevance)' }, record_index: { type: 'integer', description: 'Index in the input records array' }, score: { type: 'number', description: 'Relevance score 0.0 - 1.0 (query terms matched / total terms)' }, evidence: { type: 'string', description: 'First 240 characters of record content for context' } } }, description: 'Ranked list of records matching the query, sorted by relevance descending' }, failures: { type: 'array', items: { type: 'object', description: 'Records with status="failed" or status="error"' }, description: 'Unranked list of records with failed/error status' }, replayable: { type: 'boolean', description: 'True if records list is non-empty (indicates replay capability)' }, count: { type: 'integer', description: 'Number of ranked matches' } }